CN108776677A - Creation method, equipment and the computer readable storage medium of parallel statement library - Google Patents
Creation method, equipment and the computer readable storage medium of parallel statement library Download PDFInfo
- Publication number
- CN108776677A CN108776677A CN201810526260.0A CN201810526260A CN108776677A CN 108776677 A CN108776677 A CN 108776677A CN 201810526260 A CN201810526260 A CN 201810526260A CN 108776677 A CN108776677 A CN 108776677A
- Authority
- CN
- China
- Prior art keywords
- sentence
- parallel
- question sentence
- statement library
- main clause
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of creation method, equipment and the computer readable storage mediums of parallel statement library, and the method comprising the steps of:After detecting the establishment for creating parallel statement library instruction, the effective question sentence for obtaining user's triggering in preset duration is instructed according to creating;The term vector for calculating each word in effective question sentence determines the sentence vector of effective question sentence according to term vector;Effective question sentence is clustered according to sentence vector, with the corresponding theme class of each effective question sentence of determination;The corresponding theme class of effective question sentence is added in preset parallel statement library.The present invention realizes automatically creating for parallel statement library, improve the formation efficiency of parallel sentence, and using the question sentence that real user triggers as the basis for creating parallel statement library, the parallel sentence in created parallel statement library is set to meet the question sentence pattern of real user, the otherness between parallel sentence and real user question sentence pattern is reduced, the accuracy rate for answering customer problem by knowledge base is improved.
Description
Technical field
The present invention relates to natural language processing technique field more particularly to a kind of creation method of parallel statement library, equipment
And computer readable storage medium.
Background technology
Parallel sentence can be applicable to machine translation field, and parallel sentence can be used between different language, with multiple and different languages
Kind sentence expresses identical semanteme;Parallel sentence can be also used under same languages, identical with the sentence expression of multiple and different statements
Semanteme.For example, " introduction of A products ", " what A products are " and " A products are how " etc. all indicates the same semanteme, but
It is that statement is different.Due to the complexity of natural language itself, some identical semanteme might have a variety of different sentence tables
It states, these different sentence statements can create great difficulties the true intention of computer understanding sentence.Especially in intelligent answer
The standard QA (Question and Answer, problem and answer) in field, knowledge base is limited, therefore also very to the description of problem
It is limited.But different user changes greatly the way to put questions of same matter of semantics, such as the different of word, sequence difference, colloquial style etc.
It both increases to ask questions user and is mapped to the difficulty of typical problem.In order to be matched to typical problem as far as possible, need to the greatest extent may be used
It is that typical problem adds parallel sentence, description standard problem as diversified as possible mostly energy.
The method for adding parallel sentence at present for typical problem is manually to write, i.e., based on typical problem, by knowing
The experience for knowing library editorial staff, writes the sentence of different expression as much as possible.For example, typical problem is " A product introductions ", compile
Collecting the addible parallel sentence of personnel includes:" what A products are ", " how is A products ", " what is A products " etc..But people
Work write typical problem parallel sentence efficiency it is low and need rely on editorial staff experience, it is parallel so as to cause what is edited
The problem of sentence and asked real user, has differences.
Invention content
The main purpose of the present invention is to provide a kind of creation method of parallel statement library, equipment and computer-readable storages
Medium, it is intended to solve existing during creating parallel sentence, need manually to write the corresponding parallel sentence of typical problem, lead
The technology that the problem of parallel sentence inefficiency is write in cause, the parallel sentence and asked real user edited has differences is asked
Topic.
To achieve the above object, the present invention provides a kind of creation method of parallel statement library, the wound of the parallel statement library
Construction method includes step:
After detecting the establishment for creating parallel statement library instruction, is instructed according to the establishment and obtain user in preset duration
Effective question sentence of triggering;
The term vector for calculating each word in effective question sentence, according to the term vector determine the sentence of effective question sentence to
Amount;
Effective question sentence is clustered according to the sentence vector, with the corresponding theme of each effective question sentence of determination
Class;
The corresponding theme class of the effective question sentence is added in preset parallel statement library.
Preferably, it is described by the corresponding theme class of the effective question sentence be added to the step in preset parallel statement library it
Afterwards, further include:
After detecting the inquiry instruction for inquiring the parallel statement library, preset standard is obtained according to the inquiry instruction
Question sentence;
Parallel sentence to be recommended corresponding with the standard question sentence is searched in the parallel statement library;
By the parallel sentence to be recommended include in display interface, and detect whether to detect confirm it is described to be recommended flat
Line statement is the confirmation instruction of the parallel sentence of the standard question sentence;
If detecting the confirmation instruction, the master parallel sentence to be recommended being added to where the standard question sentence
It inscribes in class.
Preferably, described that parallel sentence to be recommended corresponding with the standard question sentence is searched in the parallel statement library
Step includes:
Calculate the similarity between the standard question sentence and each theme class main clause in the parallel statement library;
Theme class corresponding more than the main clause of the first predetermined threshold value with the standard Question sentence parsing is obtained, target master is denoted as
Inscribe class;
It obtains the similarity in the target topic class between the standard question sentence and is more than the parallel of the second predetermined threshold value
Sentence obtains parallel sentence to be recommended.
Preferably, it is default to be more than second for the similarity obtained in the target topic class between the standard question sentence
The parallel sentence of threshold value further includes before the step of obtaining parallel sentence to be recommended:
It determines in the target topic class it has been recommended that the parallel sentence crossed, the parallel sentence for carrying default markup information;
The similarity obtained in the target topic class between the standard question sentence is more than the second predetermined threshold value
Parallel sentence, the step of obtaining parallel sentence to be recommended include:
Obtain in the target topic class except it is described it has been recommended that cross parallel sentence, carry the parallel of default markup information
Outside sentence, the similarity between the standard question sentence is more than the parallel sentence of the second predetermined threshold value, obtains parallel language to be recommended
Sentence.
Preferably, described that the corresponding theme class of the effective question sentence is added to the step packet in preset parallel statement library
It includes:
It determines that effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause, and obtain preset parallel sentence
The main clause of each theme class in library, is denoted as the second main clause;
The similarity between first main clause and second main clause is calculated, and obtains the maximum in the similarity
Value;
If the maximum value is more than or equal to the second predetermined threshold value, by having in theme class where first main clause
It imitates in the theme class where question sentence is added to corresponding second main clause of the maximum value;
If the maximum value be less than second predetermined threshold value, using the theme class where first main clause as newly
Theme class is added in the parallel statement library.
Preferably, the step of determination effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause include:
Obtain the display frequency that effective question sentence corresponds to each effective question sentence in theme class;
It obtains and shows the highest effective question sentence of frequency in each theme class, the highest effective question sentence of the display frequency is remembered
For the first main clause of corresponding theme class.
Preferably, the step of similarity calculated between first main clause and second main clause includes:
First main clause and second main clause are converted into corresponding primary vector sequence and secondary vector sequence;
By in the primary vector sequence and the secondary vector sequence inputting to preset two-way shot and long term memory network,
To obtain the similarity between first main clause and second main clause.
Preferably, the term vector for calculating each word in effective question sentence, has according to described in term vector determination
The step of sentence vector for imitating question sentence includes:
The term vector for calculating each word in effective question sentence, determines according to the term vector in each effective question sentence
The term vector mean value or term vector intermediate value of all words;
Using the term vector mean value or the term vector intermediate value as the sentence of effective question sentence vector.
In addition, to achieve the above object, the present invention also provides a kind of establishment equipment of parallel statement library, the parallel sentences
The establishment equipment in library includes memory, processor and is stored on the memory and can run on the processor parallel
It is realized when the establishment program of the establishment program of statement library, the parallel statement library is executed by the processor as described above parallel
The step of creation method of statement library.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer readable storage medium
The establishment program of parallel statement library is stored on storage medium, it is real when the establishment program of the parallel statement library is executed by processor
Now the step of creation method of parallel statement library as described above.
The present invention clusters effective question sentence, determines and each effectively ask by obtaining effective question sentence in preset duration
The corresponding theme class of sentence, the corresponding theme class of effective question sentence is added in parallel statement library, realizes oneself of parallel statement library
It is dynamic to create, the parallel statement library of manual creation is not needed, the formation efficiency of parallel sentence, and asking with real user triggering are improved
Sentence makes the parallel sentence in created parallel statement library meet the question sentence mould of real user as the basis for creating parallel statement library
Formula reduces the otherness between parallel sentence and real user question sentence pattern, improves and answer customer problem by knowledge base
Accuracy rate.
Description of the drawings
Fig. 1 is the structural schematic diagram for the hardware running environment that the embodiment of the present invention is related to;
Fig. 2 is the flow diagram of the creation method first embodiment of the parallel statement library of the present invention;
Fig. 3 is the flow diagram of the creation method second embodiment of the parallel statement library of the present invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific implementation mode
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
As shown in Figure 1, the structural schematic diagram for the hardware running environment that Fig. 1, which is the embodiment of the present invention, to be related to.
It should be noted that Fig. 1 can be the structural schematic diagram of the hardware running environment of the establishment equipment of parallel statement library.
The establishment equipment of the parallel statement library of the embodiment of the present invention can be PC, the terminal devices such as pocket computer.
As shown in Figure 1, the establishment equipment of the parallel statement library may include:Processor 1001, such as CPU, network interface
1004, user interface 1003, memory 1005, communication bus 1002.Wherein, communication bus 1002 for realizing these components it
Between connection communication.User interface 1003 may include display screen (Display), input unit such as keyboard (Keyboard),
Optional user interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 may include optionally
Standard wireline interface and wireless interface (such as WI-FI interfaces).Memory 1005 can be high-speed RAM memory, can also be steady
Fixed memory (non-volatile memory), such as magnetic disk storage.Memory 1005 optionally can also be independently of
The storage device of aforementioned processor 1001.
It will be understood by those skilled in the art that the establishment device structure of parallel statement library shown in Fig. 1 is not constituted pair
The restriction of the establishment equipment of parallel statement library may include components more more or fewer than diagram, or combine certain components, or
The different component arrangement of person.
As shown in Figure 1, as may include that operating system, network are logical in a kind of memory 1005 of computer storage media
Believe module, the establishment program of Subscriber Interface Module SIM and parallel statement library.Wherein, operating system is management sentence parallel with controlling
The program of the establishment device hardware and software resource in library supports the establishment program and other softwares or program of parallel statement library
Operation.
In the establishment equipment of parallel statement library shown in Fig. 1, user interface 1003 is mainly used for obtaining input by user
Effective question sentence, operational order and output prompt message etc.;Network interface 1004 is mainly used for connecting background server, and rear
Platform server is into row data communication, as searched the corresponding answer that waits answering a question;And processor 1001 can be used for calling storage
The establishment program of the parallel statement library stored in device 1005, and execute following operation:
After detecting the establishment for creating parallel statement library instruction, is instructed according to the establishment and obtain user in preset duration
Effective question sentence of triggering;
The term vector for calculating each word in effective question sentence, according to the term vector determine the sentence of effective question sentence to
Amount;
Effective question sentence is clustered according to the sentence vector, with the corresponding theme of each effective question sentence of determination
Class;
The corresponding theme class of the effective question sentence is added in preset parallel statement library.
Further, described that the corresponding theme class of the effective question sentence is added to the step in preset parallel statement library
Later, processor 1001 can be also used for calling the establishment program of parallel statement library stored in memory 1005, and execute with
Lower step:
After detecting the inquiry instruction for inquiring the parallel statement library, preset standard is obtained according to the inquiry instruction
Question sentence;
Parallel sentence to be recommended corresponding with the standard question sentence is searched in the parallel statement library;
By the parallel sentence to be recommended include in display interface, and detect whether to detect confirm it is described to be recommended flat
Line statement is the confirmation instruction of the parallel sentence of the standard question sentence;
If detecting the confirmation instruction, the master parallel sentence to be recommended being added to where the standard question sentence
It inscribes in class.
Further, described that parallel sentence to be recommended corresponding with the standard question sentence is searched in the parallel statement library
The step of include:
Calculate the similarity between the standard question sentence and each theme class main clause in the parallel statement library;
Theme class corresponding more than the main clause of the first predetermined threshold value with the standard Question sentence parsing is obtained, target master is denoted as
Inscribe class;
It obtains the similarity in the target topic class between the standard question sentence and is more than the parallel of the second predetermined threshold value
Sentence obtains parallel sentence to be recommended.
Further, it is pre- to be more than second for the similarity obtained in the target topic class between the standard question sentence
If the parallel sentence of threshold value, before the step of obtaining parallel sentence to be recommended, processor 1001 can be also used for calling memory
The establishment program of the parallel statement library stored in 1005, and execute following steps:
It determines in the target topic class it has been recommended that the parallel sentence crossed, the parallel sentence for carrying default markup information;
The similarity obtained in the target topic class between the standard question sentence is more than the second predetermined threshold value
Parallel sentence, the step of obtaining parallel sentence to be recommended include:
Obtain in the target topic class except it is described it has been recommended that cross parallel sentence, carry the parallel of default markup information
Outside sentence, the similarity between the standard question sentence is more than the parallel sentence of the second predetermined threshold value, obtains parallel language to be recommended
Sentence.
Further, described that the corresponding theme class of the effective question sentence is added to the step in preset parallel statement library
Including:
It determines that effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause, and obtain preset parallel sentence
The main clause of each theme class in library, is denoted as the second main clause;
The similarity between first main clause and second main clause is calculated, and obtains the maximum in the similarity
Value;
If the maximum value is more than or equal to the second predetermined threshold value, by having in theme class where first main clause
It imitates in the theme class where question sentence is added to corresponding second main clause of the maximum value;
If the maximum value be less than second predetermined threshold value, using the theme class where first main clause as newly
Theme class is added in the parallel statement library.
Further, the step of determination effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause is wrapped
It includes:
Obtain the display frequency that effective question sentence corresponds to each effective question sentence in theme class;
It obtains and shows the highest effective question sentence of frequency in each theme class, the highest effective question sentence of the display frequency is remembered
For the first main clause of corresponding theme class.
Further, the step of similarity calculated between first main clause and second main clause includes:
First main clause and second main clause are converted into corresponding primary vector sequence and secondary vector sequence;
By in the primary vector sequence and the secondary vector sequence inputting to preset two-way shot and long term memory network,
To obtain the similarity between first main clause and second main clause.
Further, the term vector for calculating each word in effective question sentence, according to term vector determination
The step of the sentence vector of effective question sentence includes:
The term vector for calculating each word in effective question sentence, determines according to the term vector in each effective question sentence
The term vector mean value or term vector intermediate value of all words;
Using the term vector mean value or the term vector intermediate value as the sentence of effective question sentence vector.
Based on above-mentioned structure, each embodiment of the creation method of parallel statement library is proposed.The establishment of parallel statement library
Method is applied to the establishment equipment of parallel statement library, and the establishment equipment of parallel statement library can be PC, and the terminals such as pocket computer are set
It is standby.In order to which the simplicity of description saves the wound of slightly parallel statement library in each embodiment of the creation method of following parallel statement library
Build this standby executive agent.
With reference to Fig. 2, Fig. 2 is the flow diagram of the creation method preferred embodiment of the parallel statement library of the present invention.
An embodiment of the present invention provides the embodiments of the creation method of parallel statement library, it should be noted that although flowing
Logical order is shown in journey figure, but in some cases, it can be with different from shown or described by sequence execution herein
The step of.
The creation method of parallel statement library includes:
Step S10, after detecting the establishment for creating parallel statement library instruction, when obtaining default according to establishment instruction
Effective question sentence of user's triggering in long.
After detecting the establishment for creating parallel statement library instruction, user's triggering in preset duration is obtained according to creating to instruct
Effective question sentence.Wherein, by the establishment equipment of parallel statement library according to timed task clocked flip, which appoints for establishment instruction
Corresponding duration of being engaged in can be arranged according to specific needs.It such as may be configured as at interval of 10 days, 15 days or 30 days.Preset duration
It is consistent to may be configured as duration corresponding with timed task, or is set as inconsistent with timed task.Effective question sentence is to meet to ask
The problem of question sentence of sentence condition, effective question sentence is asked by user in preset duration in intelligent Answer System.Wherein, intelligent answer
System can be applied in the establishment of parallel statement library, it can answer the problem of user is carried automatically.Intelligent Answer System includes
Parallel statement library and knowledge base are stored with standard question sentence in knowledge base.
The process of effectively question sentence is in acquisition preset duration:The question sentence in preset duration in database is obtained, judges to be obtained
Whether within a preset range the number of characters of the question sentence taken, and judges whether acquired question sentence is number entirely and/or is entirely
Letter.If the number of characters of acquired question sentence is within a preset range, and acquired question sentence does not contain number and/or letter, or
Person contains only part number and/or subalphbet, it is determined that acquired question sentence is effective question sentence.If the word of acquired question sentence
Accord with number not within a preset range and/or acquired question sentence be entirely letter and/or number, it is determined that acquired question sentence
For invalid question sentence.Wherein, preset range can be arranged according to specific needs, such as may be configured as 6 characters to 40 characters, or
8 characters are set as to 50 characters etc..
Further, it can also judge that acquired question sentence whether there is forbidden character, such as judge that acquired question sentence is
It is no that there are dirty words.If there are dirty words for acquired question sentence, it is determined that acquired question sentence is invalid question sentence.
Further, it in order to reduce the calculation amount for the term vector for calculating effective question sentence, after getting effective question sentence, deletes
The garbages such as space, digital number in effective question sentence.As being " 1, A products how " when some effective question sentence, then delete
In effective question sentence number " 1 " and symbol ", ".
Step S20 calculates the term vector of each word in effective question sentence, is determined and described is effectively asked according to the term vector
The sentence vector of sentence.
After getting effective question sentence, effective question sentence is segmented, obtains each word in effective question sentence, and calculating has
The term vector for imitating each word in question sentence determines the sentence vector of effective question sentence according to the term vector.Wherein, effective question sentence is divided
The segmentation methods of word include but not limited to the segmentation methods based on dictionary, the segmentation methods based on statistics and rule-based participle
Algorithm.In embodiments herein, the term vector of each word in effective question sentence is calculated by word2vec.Word2vec can be with
It is efficiently trained on the dictionary and more than one hundred million data sets of million orders of magnitude, the training knot that word2vec tools obtain
Fruit --- term vector (word embedding), can measure the similitude between word and word well.In word2vec,
Word2vec is broadly divided into CBOW (Continuous Bag of Words) and two kinds of models of Skip-Gram.CBOW is from original
Sentence speculates that target words, CBOW model equivalencies are multiplied by an embedding matrix in the vector of a bag of words, to
To a continuous embedding vector;And Skip-Gram is exactly the opposite, is to deduce original statement from target words.It can be with
Understand, in the present embodiment, language processing tools are alternatively other tools that may be implemented with word2vec said functions.
Further, step S20 includes:
Step a calculates the term vector of each word in effective question sentence, is determined according to the term vector each described effective
The term vector mean value or term vector intermediate value of all words in question sentence.
Step b, using the term vector mean value or the term vector intermediate value as the sentence of effective question sentence vector.
After calculating the term vector of each word in effective question sentence, is corresponded to according to term vector and determine institute in each effective question sentence
There are the term vector mean value or term vector intermediate value of word.It, will be effective in the effective question sentence of determination during the term vector mean value of all words
The corresponding value of the term vector of all words is added in question sentence, and the quantity of word, correspondence obtain term vector then divided by effective question sentence
Mean value.In the effective question sentence of determination during the term vector intermediate value of all words, by the mould of all term vectors in effective question sentence according to
Sequence sequence from small to large, takes the mould for being arranged in centre to correspond to term vector as term vector intermediate value.
After the term vector mean value or term vector intermediate value of all words in the effective question sentence of determination, by term vector mean value or word to
Amount intermediate value is vectorial to the sentence that should be used as effective question sentence.Further, the maximum of all term vector moulds in effective question sentence can also be obtained
Value or minimum value, the maximum value of term vector mould or the corresponding term vector of minimum value is vectorial to the sentence that should be used as effective question sentence.
Such as when some effective question sentence has 7 words, corresponding term vector is respectively WithIfWhen (in calculating process, carried out with each vector field homoemorphism
Calculate), then it can incite somebody to actionSentence vector as effective question sentence.If suitable according to from small to large according to the term vector mould of this 7 words
After sequence sequence, gained ranking results are: WithThen determineFor effective question sentence sentence to
Amount.
Step S30 clusters effective question sentence according to the sentence vector, with each effective question sentence pair of determination
The theme class answered.
After the sentence vector of the effective question sentence of determination, according to the sentence vector of each effective question sentence to acquired effective question sentence into
Row cluster, obtains the corresponding cluster of each effective question sentence, with the corresponding theme class of each effective question sentence of determination.It is understood that
One cluster is a theme class, and effective question sentence in each cluster expresses the same semanteme.It is exactly pair to be clustered to effective question sentence
Effective question sentence is classified, i.e., acquired effective question sentence is classified according to semanteme, will express effectively asking for same semanteme
Sentence is divided into same class.It should be noted that between any two cluster, i.e., the cluster centre between any two theme class is similar
Degree should be less than predetermined threshold value, and cluster centre is that effective question sentence as cluster benchmark, cluster centre are corresponding in corresponding theme class
Effective question sentence can be pre-set according to specific needs.Predetermined threshold value can be arranged according to specific needs, in the present embodiment not
Do concrete restriction.
Wherein, acquired effective question sentence is carried out clustering clustering algorithm used according to the sentence vector of each effective question sentence
Including but not limited to K-Means (K mean values) cluster, mean shift clustering, density clustering method (DBSCAN) and with height
This mixed model (GMM, Adaptive background mixture models for real-time tracking) is most
It is big it is expected (EM, Expectation Maximization Algorithm) cluster.
The corresponding theme class of the effective question sentence is added in preset parallel statement library by step S40.
After determining acquired effective question sentence corresponding theme class, the corresponding theme class of effective question sentence is added to default
Parallel statement library in.It is understood that be added in preset parallel statement library by the corresponding theme class of effective question sentence,
It is to be added to effective question sentence included in the theme class in preset parallel statement library as a theme class.Preset
In parallel statement library, the parallel sentence of different themes class is had existed.Parallel statement library is by the corresponding theme of parallel sentence
Class composition, a theme class includes at least two same semantemes of expression, but states different parallel sentences.
Further, step S40 includes:
Step c determines that effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause, and obtains preset flat
The main clause of each theme class in line statement library, is denoted as the second main clause.
Further, after determining acquired effective question sentence corresponding theme class, determine that effective question sentence corresponds to theme class
In main clause, and the main clause that effective question sentence is corresponded to theme class is denoted as the first main clause, and obtains in preset parallel statement library
Have the main clause of theme class, and the main clause of theme class in parallel statement library is denoted as the second main clause.
Further, the step of determination effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause is wrapped
It includes:
Step c1 obtains the display frequency that effective question sentence corresponds to each effective question sentence in theme class.
Further, it is determined that the process for the main clause that effectively question sentence corresponds to theme class is:It obtains effective question sentence and corresponds to theme class
In each effective question sentence display frequency, that is, obtain determined by theme class all effective question sentences display frequency.Wherein, it shows
Show that frequency is the number that effective question sentence is inquired in preset duration by user, i.e., when a user asks primary effective question sentence
When, 1 is added on the counter of the effective question sentence of correspondence.
Step c2 is obtained and is shown the highest effective question sentence of frequency in each theme class, is had the display frequency is highest
Effect question sentence is denoted as the first main clause of corresponding theme class.
After getting the display frequency of each effective question sentence in theme class, obtains and show frequency highest in each theme class
Effective question sentence, and will show that the highest effective question sentence of frequency is denoted as the main clause of corresponding theme class, i.e., will show that frequency is highest
Effective question sentence is denoted as the first main clause of corresponding theme class.
Further, in order to improve the accuracy rate for determining the first main clause, show frequency most in getting each theme class
After high effective question sentence, it will show that the highest effective question sentence of frequency is denoted as target main clause, and judge whether deposited in each theme class
In at least two target main clauses.If there are at least two target main clauses in some theme class, target master in the theme class is obtained
Target main clause more than target main clause number of words is denoted as the first main clause of the theme class by the number of words of sentence;If in each theme class all only
There are a target main clauses, then the target main clause is denoted as to the first main clause of the theme class.
Further, it during determining theme class corresponding first main clause, can also directly acquire in each theme class
Then the display frequency and number of words of effective question sentence correspond to according to the weight between display frequency and number of words and calculate each effective question sentence
Score value, effective question sentence of highest scoring is denoted as to the first main clause of corresponding theme class.Wherein, it shows between frequency and number of words
Weight can be arranged according to specific needs, can will such as show that the weight between frequency and number of words is set as 6:4, or it is set as 7:
3 etc..Different display frequencies and number of words correspond to different scores.
Step d calculates the similarity between first main clause and second main clause, and obtains in the similarity
Maximum value.
After determining the first main clause and the second main clause, the similarity between the first main clause and the second main clause is calculated, and obtain
The maximum value of similarity between first main clause and the second main clause.It should be noted that containing at least in preset parallel statement library
Two theme class, there are one main clauses for theme class, due to there are at least two theme class in parallel statement library,
At least there are two the second main clauses in embodiment.Of the number of first main clause theme class corresponding with identified effective question sentence
Number is equal.Such as when having 3 according to the theme class of acquired effective question sentence determination, then the number of the first main clause is also 3;Work as root
The theme class determined according to acquired effective question sentence has 5, then the number of the first main clause is also 5.
Further, step d includes:
First main clause and second main clause are converted into corresponding primary vector sequence and secondary vector by step d1
Sequence.
Further, the process of similarity is between the first main clause of calculating and the second main clause:First main clause is changed into first
Sequence vector, and the second main clause is changed into secondary vector sequence.Specifically, can by word2vec tools by the first main clause and
Second main clause is converted into corresponding primary vector sequence and secondary vector sequence.By word2vec tools by the first main clause and
Second main clause is converted into corresponding primary vector sequence and secondary vector sequence process, can be according to the first main clause and the second main clause
In each word put in order, the term vector of each word is arranged, corresponding primary vector sequence and secondary vector is obtained
Sequence.
Step d2 remembers the primary vector sequence and the secondary vector sequence inputting to preset two-way shot and long term
In network, to obtain the similarity between first main clause and second main clause.
After obtaining primary vector sequence and secondary vector sequence, extremely by primary vector sequence and secondary vector sequence inputting
In preset two-way shot and long term memory network (LSTM, Long Short-Term Memory), two-way shot and long term memory net is obtained
The value of network output.It is understood that the value of two-way shot and long term memory network output is between the first main clause and the second main clause
Similarity.Wherein, two-way shot and long term memory network is pre-created.Creating two-way shot and long term memory network process
In, obtain similar corpus.In similar corpus, corresponding phase is set between each two sentence via corresponding user
Like angle value.If " today, weather was fine " and " today is fine " similarity value are 1;" today, weather was fine " and " today weather
It is very poor " similarity value be 0.In the present embodiment, ranging from the 0 to 1 of the value of two-way shot and long term memory network output, two-way length
The value of phase memory network output is higher, more similar between two sentences of expression.
After getting similar corpus, any two sentence in corpus is converted into corresponding sequence vector, it is defeated
Enter into two-way shot and long term memory network, two-way long-term short-term memory network is made to export similarity corresponding with the two sentences
Value, to build two-way shot and long term memory network.
Step e, if the maximum value is more than or equal to the second predetermined threshold value, by theme class where first main clause
In effective question sentence be added in the theme class where corresponding second main clause of the maximum value.
Step f makees the theme class where first main clause if the maximum value is less than second predetermined threshold value
It is added in the parallel statement library for new theme class.
After determining the maximum value of similarity between the first main clause and the second main clause, judge the first main clause and the second main clause it
Between similarity maximum value whether be more than or equal to the second predetermined threshold value.If similarity between the first main clause and the second main clause
Maximum value is more than or equal to the second predetermined threshold value, then effective question sentence in theme class where the first main clause is added to maximum value
In theme class where corresponding second main clause;If it is pre- to be less than second for the maximum value of similarity between the first main clause and the second main clause
If threshold value, then it is added to the theme class where the first main clause as new theme class in parallel statement library.It is understood that
It is added in parallel statement library using the theme class where the first main clause as new theme class, is by the master where the first main clause
All effective question sentences are added to as new theme class in parallel statement library in topic class.Wherein, the second predetermined threshold value can basis
It specifically needs and is arranged, be not particularly limited in the present embodiment.
Such as the first master is being calculated there are when 3 second main clause b1, b2 and b3 when there are 3 first main clauses a1, a2 and a3
After similarity between sentence and the second main clause, 9 similarities, the similarity between respectively a1 and b1, b2, b3, a2 can be obtained
The similarity between similarity and a3 and b1, b2, b3 between b1, b2, b3.If the similarity between a1 and b1, b2, b3
In, the value of similarity is maximum value between a1 and b1, is denoted as the first maximum value;In similarity between a2 and b1, b2, b3,
The value of similarity is maximum value between a2 and b3, is denoted as the second maximum value;In similarity between a3 and b1, b2, b3, a3 and
The value of similarity is maximum value between b2, is denoted as third maximum value, and the first maximum value and the second maximum value are more than or equal to
Second predetermined threshold value, third maximum value are less than the second predetermined threshold value, then add effective question sentence in the theme class a01 where a1
Into the theme class where b1, effective question sentence in the theme class a02 where a2 is added in the theme class where b3, by a3
The theme class a03 at place is as theme class new in parallel statement library.
Further, after determining each effective question sentence corresponding theme class, can not also consider in parallel statement library
Identified theme class is directly added in parallel statement library by some theme class.
The present embodiment clusters effective question sentence, determines each effective by obtaining effective question sentence in preset duration
The corresponding theme class of effective question sentence is added in parallel statement library, realizes parallel statement library by the corresponding theme class of question sentence
It automatically creates, does not need the parallel statement library of manual creation, improve the formation efficiency of parallel sentence, and with real user triggering
Question sentence makes the parallel sentence in created parallel statement library meet the question sentence of real user as the basis for creating parallel statement library
Pattern reduces the otherness between parallel sentence and real user question sentence pattern, improves and is asked by knowledge base answer user
The accuracy rate of topic.
Further, the creation method second embodiment of the parallel statement library of the present invention is proposed.
The creation method second embodiment of the parallel statement library is implemented with the creation method first of the parallel statement library
Difference lies in reference to Fig. 3, the creation method of parallel statement library further includes example:
Step S50 is obtained pre- after detecting the inquiry instruction for inquiring the parallel statement library according to the inquiry instruction
If standard question sentence.
In real time or timing detects whether to detect the inquiry instruction for inquiring parallel statement library.Parallel language is inquired when detecting
After the inquiry instruction in sentence library, preset standard question sentence is obtained according to the inquiry instruction.Wherein, standard question sentence can be according to specific needs
And be arranged, one or more standard question sentence may be present in a theme class.
Step S60 searches parallel sentence to be recommended corresponding with the standard question sentence in the parallel statement library.
After getting standard question sentence, parallel sentence to be recommended corresponding with standard question sentence is searched in parallel statement library.
It is understood that after getting standard question sentence, the theme class where standard question sentence is determined, by the parallel language in the theme class
Sentence is used as the corresponding parallel sentence to be recommended of standard question sentence.
Further, step S60 includes:
Step h is calculated similar between the standard question sentence and each theme class main clause in the parallel statement library
Degree.
Further, the process of lookup parallel sentence to be recommended corresponding with standard question sentence can be in parallel statement library:
The main clause of each theme class in parallel statement library is obtained, and calculates the main clause of standard question sentence and each theme class in parallel statement library
Between similarity.The algorithm used in similarity between calculating standard question sentence and the main clause of each theme class in parallel statement library
Similarity algorithm between the first main clause of calculating and the second main clause is consistent, repeats no more in the present embodiment.
During the main clause of theme class in obtaining parallel statement library, it can determine whether each in each theme class in parallel statement library
Whether a parallel sentence carries main clause mark.If some parallel sentence carries main clause mark, it is determined that carry the flat of main clause mark
Line statement is the main clause of corresponding theme class.Wherein, the specific manifestation form of main clause mark can be arranged according to specific needs, at this
It is not limited in embodiment.
Step i obtains theme class corresponding more than the main clause of the first predetermined threshold value with the standard Question sentence parsing, is denoted as
Target topic class.
After calculating the similarity between standard question sentence and each theme class main clause in parallel statement library, each master is judged
Inscribe whether the similarity between class main clause and standard question sentence is more than the first predetermined threshold value, if the main clause of theme class and standard question sentence it
Between similarity be more than the first predetermined threshold value, then obtain and standard question sentence between similarity be more than the first predetermined threshold value main clause pair
Similarity between standard question sentence theme class corresponding more than the main clause of the first predetermined threshold value is denoted as target by the theme class answered
Theme class.Wherein, the first predetermined threshold value can be arranged according to specific needs, and the first predetermined threshold value can be with the second predetermined threshold value phase
Deng also can be unequal with the second predetermined threshold value.It is understood that target topic class can be one or more.
Step j obtains the similarity in the target topic class between the standard question sentence and is more than the second predetermined threshold value
Parallel sentence, obtain parallel sentence to be recommended.
After determining target topic class, the parallel sentence in target topic class is obtained, and calculate parallel in target topic class
Similarity between sentence and standard question sentence, and determine in target topic class, the similarity between parallel sentence and standard question sentence
Parallel sentence corresponding more than the second predetermined threshold value, obtains parallel sentence to be recommended.It should be noted that between standard question sentence
Similarity be more than the second predetermined threshold value parallel sentence be parallel sentence to be recommended.Calculating standard question sentence and target topic class
Parallel sentence between similarity used in algorithm it is consistent with similarity algorithm between the first main clause and the second main clause is calculated,
It repeats no more in the present embodiment.
As if it is determined that target topic class is P, when standard question sentence is Q, if there are P1, P2, P3, P4, P5 in target topic class
With this 6 parallel sentences of P6, but the similarity only between P1, P3 and P4 and Q be more than the second predetermined threshold value, it is determined that P1, P3
It is parallel sentence to be recommended with P4.
The parallel sentence to be recommended is included and detecting whether to detect described in confirmation in display interface by step S70
Parallel sentence to be recommended is the confirmation instruction of the parallel sentence of the standard question sentence.
Include and detecting whether to detect in display interface by parallel sentence to be recommended after obtaining parallel sentence to be recommended
Measure the confirmation instruction for the parallel sentence for confirming that parallel sentence to be recommended is standard question sentence.Wherein, confirmation instruction is by triggering
The corresponding user of inquiry instruction is triggered.After including in display interface by parallel sentence to be recommended, prompt message is exported,
With according to prompt message prompt confirm parallel sentence to be recommended whether be standard question sentence parallel sentence.It such as can be in display interface
The corresponding button of middle display " receiving " and " not receiving ", when detecting the clicking operation for clicking " receiving " button, determination waits pushing away
Recommend the parallel sentence that parallel sentence is standard question sentence;When detecting the clicking operation for clicking " not receiving " button, determination waits pushing away
It is the parallel sentence of standard question sentence to recommend parallel sentence not.
Further, the parallel sentence of standard question sentence is quickly found for the ease of inquiry user, will be to be recommended parallel
Sentence is shown in during display interface, is calculated the similarity between each parallel sentence to be recommended and standard question sentence, will be waited pushing away
Parallel sentence is recommended according to the size of similarity between each parallel sentence to be recommended and standard question sentence, is shown in display from big to small
In interface.If in the presence of parallel sentence to be recommended identical with standard Question sentence parsing, obtain identical as standard Question sentence parsing
Parallel sentence to be recommended display frequency within some period, will show that the big parallel sentence to be recommended of frequency includes aobvious
Before showing the small parallel sentence to be recommended of frequency.Further, detect whether parallel sentence to be recommended carries default mark.If waiting for
Recommend parallel sentence to carry default mark, then shows the parallel sentence to be recommended for carrying default mark up front;If not
In the presence of the parallel sentence to be recommended for carrying default mark, then by parallel sentence to be recommended according to each parallel sentence to be recommended and mark
The size of similarity, is shown in display interface from big to small between quasi- question sentence.Wherein, if some parallel sentence to be recommended carries
There is default mark, then it is " identical " to show that the parallel sentence to be recommended is marked user annotation, indicates the parallel sentence to be recommended
It is identical as the main clause semanteme of target topic class where it.The specific manifestation form of default mark can be set according to specific needs
It sets, is not limited in the present embodiment.
The parallel sentence to be recommended is added to the standard question sentence by step S80 if detecting the confirmation instruction
In the theme class at place.
If detecting the confirmation instruction for the parallel sentence for confirming that parallel sentence to be recommended is standard question sentence, instruction will confirm that
Corresponding parallel sentence to be recommended is added in the theme class where standard question sentence;Confirm parallel sentence to be recommended if not detecting
For the confirmation instruction of the parallel sentence of standard question sentence, then it not will confirm that instruction is corresponding and parallel sentence recommended to be added to standard question sentence
In the theme class at place.
The present embodiment is determined by after detecting inquiry instruction, obtaining preset standard question sentence in parallel statement library
The corresponding parallel sentence of the standard question sentence, to quickly generate the corresponding parallel sentence of standard question sentence.
Further, the creation method 3rd embodiment of the parallel statement library of the present invention is proposed.
The creation method the first or the of the creation method 3rd embodiment of the parallel statement library and the parallel statement library
Difference lies in the creation method of parallel statement library further includes two embodiments:
Step k, determine in the target topic class it has been recommended that cross parallel sentence, carry the parallel of default markup information
Sentence.
After determining target topic class, determine in target topic class it has been recommended that cross parallel sentence, carry default mark
The parallel sentence of information.In the present embodiment, it has been recommended that the parallel sentence crossed, carrying the parallel sentence point of default markup information
It is not indicated with different identification informations.Such as in target topic class, if some parallel sentence carries the identification information of " 01 ",
Determine that the parallel sentence is it has been recommended that the parallel sentence crossed;If some parallel sentence carries the identification information of " 11 ", it is determined that should
Parallel sentence is to carry the parallel sentence for presetting markup information.Wherein, it is to be corresponded to carry and preset the parallel sentence of markup information
Mark user annotation be " difference " parallel sentence, indicate the parallel sentence and its where theme class main clause semanteme not
Together.
Step j includes:
Step j1, obtain in the target topic class except it is described it has been recommended that cross parallel sentence, carry default mark letter
Outside the parallel sentence of breath, the similarity between the standard question sentence is more than the parallel sentence of the second predetermined threshold value, obtains waiting pushing away
Recommend parallel sentence.
When determine in target topic class it has been recommended that cross parallel sentence, carry the parallel sentence of default markup information after,
Obtain target topic class in except it has been recommended that cross parallel sentence, carry the parallel sentence of default markup information in addition to, asked with standard
Similarity is more than the parallel sentence of the second predetermined threshold value between sentence, similarity will be more than the second predetermined threshold value between standard question sentence
Parallel sentence as parallel sentence to be recommended.
The present embodiment is by before obtaining parallel sentence to be recommended, determining in target topic class it has been recommended that the parallel language crossed
Sentence, the parallel sentence for carrying default markup information do not consider during determining parallel sentence to be recommended in target topic class
It has been recommended that the parallel sentence crossed, carrying the parallel sentence of default markup information, parallel sentence process to be recommended is obtained to reduce
In calculation amount, improve the speed for determining parallel sentence to be recommended.
In addition, the embodiment of the present invention also proposes a kind of computer readable storage medium, the computer readable storage medium
On be stored with the establishment program of parallel statement library, the establishment program of the parallel statement library realizes institute as above when being executed by processor
The step of reward sending method stated.
Computer readable storage medium specific implementation mode of the present invention is respectively implemented with the creation method of above-mentioned parallel statement library
Example is essentially identical, and details are not described herein.
It should be noted that herein, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that process, method, article or device including a series of elements include not only those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including this
There is also other identical elements in the process of element, method, article or device.
The embodiments of the present invention are for illustration only, can not represent the quality of embodiment.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical scheme of the present invention substantially in other words does the prior art
Going out the part of contribution can be expressed in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disc, CD), including some instructions are used so that a station terminal equipment (can be mobile phone, computer, clothes
Be engaged in device, air conditioner or the network equipment etc.) execute method described in each embodiment of the present invention.
It these are only the preferred embodiment of the present invention, be not intended to limit the scope of the invention, it is every to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of creation method of parallel statement library, which is characterized in that the creation method of the parallel statement library includes following step
Suddenly:
After detecting the establishment for creating parallel statement library instruction, is instructed according to the establishment and obtain user's triggering in preset duration
Effective question sentence;
The term vector for calculating each word in effective question sentence determines the sentence vector of effective question sentence according to the term vector;
Effective question sentence is clustered according to the sentence vector, with the corresponding theme class of each effective question sentence of determination;
The corresponding theme class of the effective question sentence is added in preset parallel statement library.
2. the creation method of parallel statement library as described in claim 1, which is characterized in that described to correspond to effective question sentence
Theme class be added to after the step in preset parallel statement library, further include:
After detecting the inquiry instruction for inquiring the parallel statement library, preset standard is obtained according to the inquiry instruction and is asked
Sentence;
Parallel sentence to be recommended corresponding with the standard question sentence is searched in the parallel statement library;
Include and detecting whether to detect and confirming the parallel language to be recommended in display interface by the parallel sentence to be recommended
Sentence is the confirmation instruction of the parallel sentence of the standard question sentence;
If detecting the confirmation instruction, the theme class parallel sentence to be recommended being added to where the standard question sentence
In.
3. the creation method of parallel statement library as claimed in claim 2, which is characterized in that described in the parallel statement library
Search parallel sentence to be recommended corresponding with the standard question sentence the step of include:
Calculate the similarity between the standard question sentence and each theme class main clause in the parallel statement library;
Theme class corresponding more than the main clause of the first predetermined threshold value with the standard Question sentence parsing is obtained, target topic is denoted as
Class;
The parallel sentence that the similarity in the target topic class between the standard question sentence is more than the second predetermined threshold value is obtained,
Obtain parallel sentence to be recommended.
4. the creation method of parallel statement library as claimed in claim 3, which is characterized in that described to obtain the target topic class
In similarity between the standard question sentence be more than the parallel sentence of the second predetermined threshold value, obtain the step of parallel sentence to be recommended
Before rapid, further include:
It determines in the target topic class it has been recommended that the parallel sentence crossed, the parallel sentence for carrying default markup information;
The similarity obtained in the target topic class between the standard question sentence is more than the parallel of the second predetermined threshold value
Sentence, the step of obtaining parallel sentence to be recommended include:
It obtains in the target topic class except described it has been recommended that the parallel sentence crossed, the parallel sentence for carrying default markup information
Outside, the similarity between the standard question sentence is more than the parallel sentence of the second predetermined threshold value, obtains parallel sentence to be recommended.
5. the creation method of parallel statement library as described in claim 1, which is characterized in that described to correspond to effective question sentence
The step that is added in preset parallel statement library of theme class include:
It determines that effective question sentence corresponds to the main clause in theme class, is denoted as the first main clause, and obtain in preset parallel statement library
The main clause of each theme class is denoted as the second main clause;
The similarity between first main clause and second main clause is calculated, and obtains the maximum value in the similarity;
If the maximum value is more than or equal to the second predetermined threshold value, by effectively asking in theme class where first main clause
Sentence is added in the theme class where corresponding second main clause of the maximum value;
If the maximum value is less than second predetermined threshold value, using the theme class where first main clause as new theme
Class is added in the parallel statement library.
6. the creation method of parallel statement library as claimed in claim 5, which is characterized in that determination effective question sentence pair
The step of answering the main clause in theme class, being denoted as the first main clause include:
Obtain the display frequency that effective question sentence corresponds to each effective question sentence in theme class;
It obtains and shows the highest effective question sentence of frequency in each theme class, the highest effective question sentence of display frequency is denoted as pair
Answer the first main clause of theme class.
7. the creation method of parallel statement library as claimed in claim 5, which is characterized in that it is described calculate first main clause and
The step of similarity between second main clause includes:
First main clause and second main clause are converted into corresponding primary vector sequence and secondary vector sequence;
By in the primary vector sequence and the secondary vector sequence inputting to preset two-way shot and long term memory network, with
Similarity between first main clause and second main clause.
8. the creation method of parallel statement library as described in any one of claim 1 to 7, which is characterized in that described in the calculating
The term vector of each word, determines that the step of the sentence vector of effective question sentence includes according to the term vector in effective question sentence:
The term vector for calculating each word in effective question sentence determines in each effective question sentence own according to the term vector
The term vector mean value or term vector intermediate value of word;
Using the term vector mean value or the term vector intermediate value as the sentence of effective question sentence vector.
9. a kind of establishment equipment of parallel statement library, which is characterized in that the establishment equipment of the parallel statement library include memory,
Processor and the establishment program for being stored in the parallel statement library that can be run on the memory and on the processor, it is described flat
Such as parallel sentence described in any item of the claim 1 to 8 is realized when the establishment program in line statement library is executed by the processor
The step of creation method in library.
10. a kind of computer readable storage medium, which is characterized in that be stored with parallel language on the computer readable storage medium
The establishment program in sentence library is realized when the establishment program of the parallel statement library is executed by processor as any in claim 1 to 8
The step of creation method of parallel statement library described in.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810526260.0A CN108776677B (en) | 2018-05-28 | 2018-05-28 | Parallel sentence library creating method and device and computer readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810526260.0A CN108776677B (en) | 2018-05-28 | 2018-05-28 | Parallel sentence library creating method and device and computer readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108776677A true CN108776677A (en) | 2018-11-09 |
CN108776677B CN108776677B (en) | 2021-11-12 |
Family
ID=64027910
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810526260.0A Active CN108776677B (en) | 2018-05-28 | 2018-05-28 | Parallel sentence library creating method and device and computer readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108776677B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110502752A (en) * | 2019-08-21 | 2019-11-26 | 北京一链数云科技有限公司 | A kind of text handling method, device, equipment and computer storage medium |
CN110968674A (en) * | 2019-12-04 | 2020-04-07 | 电子科技大学 | Method for constructing question comment pairs based on word vector representation |
CN111177350A (en) * | 2019-12-20 | 2020-05-19 | 北京淇瑀信息科技有限公司 | Method, device and system for forming dialect of intelligent voice robot |
CN112541352A (en) * | 2020-12-23 | 2021-03-23 | 上海永骁智能技术有限公司 | Policy interpretation method based on deep learning |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040205457A1 (en) * | 2001-10-31 | 2004-10-14 | International Business Machines Corporation | Automatically summarising topics in a collection of electronic documents |
US9400781B1 (en) * | 2016-02-08 | 2016-07-26 | International Business Machines Corporation | Automatic cognate detection in a computer-assisted language learning system |
CN106202068A (en) * | 2016-07-25 | 2016-12-07 | 哈尔滨工业大学 | The machine translation method of semantic vector based on multi-lingual parallel corpora |
CN106570171A (en) * | 2016-11-03 | 2017-04-19 | 中国电子科技集团公司第二十八研究所 | Semantics-based sci-tech information processing method and system |
US20170147691A1 (en) * | 2015-11-20 | 2017-05-25 | Guangzhou Shenma Mobile Information Technology Co. Ltd. | Method and apparatus for extracting topic sentences of webpages |
CN107391501A (en) * | 2017-09-11 | 2017-11-24 | 南京大学 | A kind of neural machine translation method of word-based prediction |
CN107545036A (en) * | 2017-07-28 | 2018-01-05 | 深圳前海微众银行股份有限公司 | Customer service robot Knowledge Database method, customer service robot and readable storage medium storing program for executing |
CN107944027A (en) * | 2017-12-12 | 2018-04-20 | 苏州思必驰信息科技有限公司 | Create the method and system of semantic key index |
CN108038234A (en) * | 2017-12-26 | 2018-05-15 | 众安信息技术服务有限公司 | A kind of question sentence template automatic generation method and device |
CN108052659A (en) * | 2017-12-28 | 2018-05-18 | 北京百度网讯科技有限公司 | Searching method, device and electronic equipment based on artificial intelligence |
-
2018
- 2018-05-28 CN CN201810526260.0A patent/CN108776677B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040205457A1 (en) * | 2001-10-31 | 2004-10-14 | International Business Machines Corporation | Automatically summarising topics in a collection of electronic documents |
US20170147691A1 (en) * | 2015-11-20 | 2017-05-25 | Guangzhou Shenma Mobile Information Technology Co. Ltd. | Method and apparatus for extracting topic sentences of webpages |
US9400781B1 (en) * | 2016-02-08 | 2016-07-26 | International Business Machines Corporation | Automatic cognate detection in a computer-assisted language learning system |
CN106202068A (en) * | 2016-07-25 | 2016-12-07 | 哈尔滨工业大学 | The machine translation method of semantic vector based on multi-lingual parallel corpora |
CN106570171A (en) * | 2016-11-03 | 2017-04-19 | 中国电子科技集团公司第二十八研究所 | Semantics-based sci-tech information processing method and system |
CN107545036A (en) * | 2017-07-28 | 2018-01-05 | 深圳前海微众银行股份有限公司 | Customer service robot Knowledge Database method, customer service robot and readable storage medium storing program for executing |
CN107391501A (en) * | 2017-09-11 | 2017-11-24 | 南京大学 | A kind of neural machine translation method of word-based prediction |
CN107944027A (en) * | 2017-12-12 | 2018-04-20 | 苏州思必驰信息科技有限公司 | Create the method and system of semantic key index |
CN108038234A (en) * | 2017-12-26 | 2018-05-15 | 众安信息技术服务有限公司 | A kind of question sentence template automatic generation method and device |
CN108052659A (en) * | 2017-12-28 | 2018-05-18 | 北京百度网讯科技有限公司 | Searching method, device and electronic equipment based on artificial intelligence |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110502752A (en) * | 2019-08-21 | 2019-11-26 | 北京一链数云科技有限公司 | A kind of text handling method, device, equipment and computer storage medium |
CN110968674A (en) * | 2019-12-04 | 2020-04-07 | 电子科技大学 | Method for constructing question comment pairs based on word vector representation |
CN110968674B (en) * | 2019-12-04 | 2023-04-18 | 电子科技大学 | Method for constructing question and comment pairs based on word vector representation |
CN111177350A (en) * | 2019-12-20 | 2020-05-19 | 北京淇瑀信息科技有限公司 | Method, device and system for forming dialect of intelligent voice robot |
CN112541352A (en) * | 2020-12-23 | 2021-03-23 | 上海永骁智能技术有限公司 | Policy interpretation method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN108776677B (en) | 2021-11-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106951494B (en) | Information recommendation method and device | |
CN108776677A (en) | Creation method, equipment and the computer readable storage medium of parallel statement library | |
CN104239501B (en) | Mass video semantic annotation method based on Spark | |
CN108628971A (en) | File classification method, text classifier and the storage medium of imbalanced data sets | |
CN105808590B (en) | Search engine implementation method, searching method and device | |
CN109299344A (en) | The generation method of order models, the sort method of search result, device and equipment | |
CN110059923A (en) | Matching process, device, equipment and the storage medium of post portrait and biographic information | |
CN108170739A (en) | Problem matching process, terminal and computer readable storage medium | |
CN109597986A (en) | Localization method, device, equipment and the storage medium of abnormal problem | |
CN107798123B (en) | Knowledge base and establishing, modifying and intelligent question and answer methods, devices and equipment thereof | |
CN106874292A (en) | Topic processing method and processing device | |
CN103678702A (en) | Video duplicate removal method and device | |
CN110364146A (en) | Audio recognition method, device, speech recognition apparatus and storage medium | |
CN110263009A (en) | Generation method, device, equipment and the readable storage medium storing program for executing of log classifying rules | |
CN112104642B (en) | Abnormal account number determination method and related device | |
CN106227714A (en) | A kind of method and apparatus obtaining the key word generating poem based on artificial intelligence | |
CN108737324A (en) | Generate the method, apparatus and relevant device, system of artificial intelligence serviced component | |
CN110134845A (en) | Project public sentiment monitoring method, device, computer equipment and storage medium | |
CN110222194A (en) | Data drawing list generation method and relevant apparatus based on natural language processing | |
CN108345612A (en) | A kind of question processing method and device, a kind of device for issue handling | |
CN111339248A (en) | Data attribute filling method, device, equipment and computer readable storage medium | |
CN111553138B (en) | Auxiliary writing method and device for standardizing content structure document | |
CN112231554A (en) | Search recommendation word generation method and device, storage medium and computer equipment | |
CN111444695A (en) | Text generation method, device and equipment based on artificial intelligence and storage medium | |
CN107545036A (en) | Customer service robot Knowledge Database method, customer service robot and readable storage medium storing program for executing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |